CAREER: Social Aggregate Measurement from Text

点击翻译按钮获取中文摘要

With rapidly expanding digitized text archives such as social media, news, and medical records, researchers are eager to measure the composition, behavior, and worldviews of diverse human populations through their linguistic expression. But it is still unknown how to construct robust and fair aggregation of text-based signals for downstream analysis. This Faculty Early Career Development project's goal is to develop fair and uncertainty-aware text analysis methods to measure socially relevant topics, sentiment, and events across bodies of text, in order to infer their relationships with variables such as author demographics, health outcomes, and political conflict. These methods will be more accurate - and be self-aware when they make mistakes - for applications such as tracking events in the news, or understanding sentiment from social media.Current methods for text aggregation are systematically biased towards idiosyncrasies in their training data, and are confounded by variability in natural language processing errors across social variables of interest. Furthermore, new probabilistic inference methods may aid insight by incorporating more richly structured natural language analysis, such as entity coreference, event parsing, or structured sentiment analysis. Through a series of case studies in the social and health sciences, this project will develop better statistical and linguistic models that combine probabilistic generative and discriminative approaches to text modeling. These methods will more accurately measure social variables' effects on topics, sentiment, and events, while inferring confidence intervals and other important statistical measures of uncertainty. Finally, this project will support education and training in interdisciplinary computational social science, through workshops at both undergraduate and graduate levels, targeting students from areas including computer science, linguistics, political science, and African-American studies.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.